| --- |
| license: mit |
| language: |
| - en |
| tags: |
| - text-classification |
| - deberta-v3 |
| - prompt-routing |
| - llm |
| - onnx |
| metrics: |
| - accuracy |
| - f1 |
| library_name: transformers |
| --- |
| |
| # DeBERTa-v3-base Prompt Complexity Classifier |
|
|
| A high-performance, lightweight prompt complexity classifier built by fine-tuning `microsoft/deberta-v3-base`. This model categorizes incoming user queries into **3 distinct complexity tiers** to power intelligent prompt routing, optimizing latency and cost in custom AI gateways. |
|
|
| ## 📊 Complexity Tiers |
|
|
| * **`TIER_0` (Low Complexity)**: Simple direct lookups, basic factual queries, typos corrections, or short conversational greetings. (Routed to fast, lightweight models). |
| * **`TIER_1` (Medium Complexity)**: Standard coding tasks, boilerplate generation, scoped bug fixes, or moderate multi-step reasoning. (Routed to balanced mid-tier models). |
| * **`TIER_2` (High Complexity)**: Advanced software architecture design, deep multi-file debugging, heavy concurrent systems analysis, or complex agentic planning. (Routed to frontier models). |
| |
| --- |
| |
| ## 🚀 Quick Start (Python) |
| |
| You can easily load and use this model for inference using the Hugging Face `transformers` library: |
| |
| ```python |
| from transformers import AutoModelForSequenceClassification, AutoTokenizer |
| import torch |
| |
| # Load model and tokenizer |
| model_id = "rohitMukhi/deberta-v3-prompt-complexity-classifier" |
| tokenizer = AutoTokenizer.from_pretrained(model_id) |
| model = AutoModelForSequenceClassification.from_pretrained(model_id) |
| |
| # Test prompt |
| prompt = "Design a multi-tenant microservices architecture with Redis clustering and strict database isolation." |
| |
| # Tokenize and predict |
| inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=256) |
| with torch.no_grad(): |
| logits = model(**inputs).logits |
| predicted_class = torch.argmax(logits, dim=1).item() |
| |
| tier_labels = {0: "TIER_0 (Low)", 1: "TIER_1 (Medium)", 2: "TIER_2 (High)"} |
| print(f"Predicted Complexity: {tier_labels[predicted_class]}") |
